Project Documentation

Project Structure

.
├── mnist-inrs/          # Dataset directory
├── src/                 # Main source code
│   ├── core/           # Core functionality
│   ├── data/           # Data handling
│   └── models/         # Model implementations
├── scripts/            # Execution scripts
│   ├── show_2.py      # Data visualization
│   ├── sanity_mlp.py  # MLP model testing
│   └── sanity_transformer.py  # Transformer testing
└── environment.yaml    # Conda environment file

Getting Started

1. Place your dataset in the mnist-inrs directory at the project root
2. Use uv to install :

   uv venv
   uv pip install -e .

Save with thisuv add your_package

Dataset Information

The project supports:

• DatasetType.MNIST: Complete MNIST dataset
• DatasetType.CIFAR10: Three selected classes (airplane, bird, automobile)

Available Scripts

Run scripts using:

python -m scripts.script_name

Current scripts:

• show_2.py: Visualize dataset
• sanity_mlp.py: Test MLP model
• sanity_transformer.py: Test Transformer model

Configuration System

The configuration system (defined in src/core/config.py) is a crucial component of this project. It provides a robust and type-safe way to configure all aspects of the experiments.

The default dataset split ratio is:

• Training: 80%
• Validation: 10%
• Test: 10%

Quick Start Configuration

For quick experimentation, use:

from src.core.config import MLPExperiment

config = MLPExperimentConfig.default()

For more detailed configuration options and parameters, refer to the docstrings in src/core/config.py.